YC W23: trends, thoughts, investments - How Agentized LLMs Will Change the Startup Landscape


Hatched by Glasp

Jul 06, 2023

4 min read


YC W23: trends, thoughts, investments - How Agentized LLMs Will Change the Startup Landscape

The Y Combinator W23 batch, which ran from January to early April 2023, was an exciting time for startups and investors alike. Each startup received a significant investment of $500K, divided into $125K at fixed terms (7%) and $375K on a SAFE MFN. This batch saw a total of $132M in invested capital for Y Combinator.

One disappointing trend that stood out was the low representation of women founders. Only 8% of the founders in this batch were women, a decrease from previous batches. Additionally, only 17% of the companies had a woman founder. These numbers are disheartening and highlight the need for more support and opportunities for women in the startup ecosystem.

In terms of industry focus, SaaS companies dominated the batch, making up 73.8% of the total. This is the highest percentage of SaaS companies in a Y Combinator batch to date. Machine Learning/AI startups were also heavily represented, with 112 companies falling into this category. Developer Tools saw a significant increase since the last batch, with 86 companies focused on this area. Fintech, Consumer, and other industries also had their presence, although to a lesser extent.

The selection process for this batch was slightly different from previous ones. Y Combinator widened the net during the initial selection process, meeting with more startups and then tightening the filter when choosing which ones to invest in. This approach allowed them to discover startups that closely aligned with their main investment interests, such as dev tools, open source, and APIs. It's interesting to note that many startups arrived at demo day with their round already full or with increased valuations, indicating a strong investor interest.

Speaking of valuations, one surprising observation is that they didn't really lower despite the market situation. Pre-seed and seed rounds remained unaffected, showing that investors still have confidence in early-stage startups. It's worth mentioning that a significant portion of the batch had no revenue (77%) or were at the idea stage (52%). This highlights the importance of early-stage funding to support the development and growth of innovative ideas.

Geographically, the Bay Area continued to be the hotspot for startups, with 158 companies based in San Francisco alone. New York, Los Angeles, and Seattle also had a notable presence. Interestingly, there were very few solo founders in this batch, with most teams consisting of 2 or 3 members. The lack of teams with 4 members may be attributed to the fact that successful startups often have founders who were already friends. This can lead to a discrepancy in gender representation, as people's best friends are likely to be of the same sex.

Pivoting is a common practice in the pre-seed stage, and this batch was no exception. At least 30% of the startups pivoted in some way during the program, demonstrating the flexibility and adaptability of founders. Many startups drew inspiration from internal tools built within big companies, developed by teams from those companies, and supported by angel investors from the same companies. This trend raises the question of whether the "Palantir Pack" is becoming the new "Paypal Mafia."

While the Y Combinator W23 batch showcased a wide range of industries and trends, it's essential to zoom out and consider the bigger picture. The emergence of agentized LLMs (Large Language Models) like Auto-GPT and Baby AGI is set to revolutionize the alignment landscape. These LLMs act as central cognitive engines, breaking tasks into subtasks, calling other software, and prioritizing and evaluating subtasks. This recursive and modular approach mirrors human intelligence, enhancing the effective intelligence of the LLM.

Integration with other approaches like HuggingGPT and Reflexion can further augment the cognitive capacities of LLMs. However, the ease of agentizing LLMs poses a challenge in terms of capabilities. In the near future, we may see an internet full of LLM-bots actively thinking and carrying out tasks. This raises concerns about alignment and coordination problems, as anyone can spawn an AGI for various purposes.

On the bright side, the ease of interpretability offered by LLMs is a significant advantage. These systems think in English, making it easier to understand their thought processes. However, it's important to note that this doesn't solve the inner alignment problem entirely. Recursive training methods may create mesa-optimizers within LLMs, which can complicate the alignment process.

In conclusion, the Y Combinator W23 batch showcased exciting trends and investments. While there are still challenges to overcome, such as improving gender representation and addressing alignment concerns with agentized LLMs, the startup ecosystem continues to evolve and innovate. Here are three actionable pieces of advice for founders and investors:

  • 1. Embrace diversity: Encourage and support women founders in the startup ecosystem. Diversity brings fresh perspectives and drives innovation.
  • 2. Stay adaptable: Pivoting is a normal part of the early-stage journey. Be open to change and willing to iterate on your ideas to find the right product-market fit.
  • 3. Prioritize alignment: As agentized LLMs become more prevalent, it's crucial to prioritize alignment and ensure that AI systems' goals align with human values. Invest in research and practices that address the alignment problem.

By keeping these insights in mind, founders and investors can navigate the ever-changing startup landscape and contribute to the growth and success of the ecosystem.

Hatch New Ideas with Glasp AI 🐣

Glasp AI allows you to hatch new ideas based on your curated content. Let's curate and create with Glasp AI :)